Why would you want to do that on Steem is EOS has sub-second transactions and is feeless too?
In 2016 VISA processed around 1,667 transactions per second (150m/day). That would at first sight make Steem an ideal blockchain for it, with a capacity of 11,000tx/second, but that would be if only very few apps would operate on the Steem blockchain. The Steem blockchain is designed to host thousands of dApps and all of those, at least the most popular ones, will create additional transactions. 11,000tx per second may not be sufficient for your thinking. Especially not when keeping in mind that the VISA network can scale to 56,000tx/second already without blockchain1).
It is true that the SWIFT network manages less transactions daily (30,5m in July 2018) but SWIFT is still growing in number of daily transactions as well. And when asking hypothetical questions, aim big.
Lastly, returning to my initial question. It seems a waste of “blockchain” to me to conceptualize/design such alternative on a blockchain without making use of the blockchain’s main asset, the unique rewards mechanic which Steem has. It isn’t because it could be done, that it should be done.
If the main focus of a platform is specific, then one should look how to make use of that asset rather than try to shoehorn other use cases in it. The main reason for that is one of capacity estimation, by which I mean that it could be that other platforms will build on Steem as well and may cause much load, create many transactions. But you could always fork and host your own witness nodes.
Yet... EOS is much faster already.
PS: Steem’s block production is three seconds, not two as stated in the question elaboration.